Market and event studies
I design research around market events, compare portfolio approaches, and test ideas with out-of-sample methods rather than relying on a single good-looking result.
Explore this work →Research · Data · AI
Jacky Xu builds useful research tools.
I use code to investigate questions, organize messy information, and turn ideas into clear, testable systems.
Selected work
AI-reinforced market analysis.
Exploring market questions through both quantitative evidence and fundamental context, with AI helping make the process faster, clearer, and easier to inspect.
Show
Research-ready data pipelines
I turn public filings, price histories, and other unstructured sources into organized inputs for analysis, with attention to timing and data quality.
Explore this work →Model evaluation workflows
I build practical ways to compare language-model outputs, inspect changes over time, and make prompt and model choices more deliberate.
Explore this work →Cognition, language, and learning
I study how language, metacognition, and self-efficacy shape learning—and how cognitive psychology can inform our understanding of machine and AI cognition.
Explore this interest →
What I do
Research systems with a human check.
I combine quantitative evidence, fundamental context, and AI-assisted workflows without hiding the reasoning behind the result.
Test market questions
Event studies, comparisons, and out-of-sample checks that separate a repeatable signal from a good story.
Keep context attached
Filings, business details, and source timing stay connected to the data used for analysis.
Make assistance inspectable
Capture prompts, sources, outputs, and review criteria so automation remains accountable.
Approach
Make the process visible.
I prefer projects that are understandable, reproducible, and honest about what the evidence can support.
Question→Evidence→System→Review
Start with a question
Define the decision or pattern worth investigating before choosing the tool.
Check the evidence
Use clean inputs, separate testing from tuning, and look for limits as well as wins.
Share clearly
Turn the work into a concise explanation someone else can inspect and use.
Evidence standards
Useful means inspectable.
The same checks guide the quantitative, fundamental, and AI sides of the work.
Keep the origin visible
Preserve links, definitions, and context so a reader can understand where an input came from.
Respect what was known
Separate information available at the time from hindsight added after the fact.
Look for the limits
Test alternatives, record failures, and say where a conclusion should not be generalized.
Current focus
Faster research, clearer judgment.
Exploring how AI can support market analysis without making the underlying evidence harder to inspect.
Privacy boundary
Enough context, no private dossier.
This site describes working interests and methods without exposing personal contact details, confidential projects, or unverifiable performance claims.
Toolkit
Tools I reach for.
These are public technical interests, not a complete profile.
Interests
Human learning, machine learning.
Cognition is a continuing area of study alongside my research, data, and AI work.
Contact